I build systems where the interesting problem is the architecture, not the CRUD — multi-agent AI pipelines with verification layers, workflow engines with real graph algorithms, and mobile apps that ship.
Most at home in Python and TypeScript, across FastAPI backends and React / React Native frontends. B.Tech in Electronics & Communication from NIT Srinagar.
Six-agent RAG pipeline that resolves e-commerce support tickets against verified order data.
An independent compliance agent re-checks every claim in a drafted reply against its cited policy chunk before it ships — a failed check triggers a rewrite loop, and a second failure escalates to a human. Knowledge base of 64 indexed chunks across 50 policy sections, evaluated on a 20-case suite.
Drag-and-drop workflow engine in the spirit of n8n and Airflow, built from scratch.
Validates graphs and resolves execution order with Kahn's topological sort, so cycles are caught before a run starts. Eight chainable node types share one base component, and the same execution engine backs both the manual trigger and an external webhook.
Asset logistics platform tracking purchases, transfers, assignments and expenditures across military bases.
Every movement recalculates net balance per asset per base, so inventory stays consistent without a reconciliation step. Three roles — Admin, Base Commander, Logistics Officer — enforced at the middleware layer rather than hidden in the UI. Deployed on Render and Vercel.
Final-year thesis benchmarking five architectures on 70,000 labelled YouTube comments.
A BERT + Bi-LSTM hybrid beat the classical baseline by 6%. Preprocessing included a hand-built 100-entry emoji dictionary, with cleaning tuned separately for classical and transformer models. Served through a live interface that pulls comments from any video link.
Whisper to BART summarization pipeline that runs entirely on-device, with no cloud APIs.
Chunks transcripts around BART's ~1024-token ceiling and stitches the results, normalizes ASR filler out of the text, and detects music segments to skip summaries that would be meaningless.
| Project | What it does | Stack |
|---|---|---|
| Network Posture Scanner | Subnet discovery and 10 CIS benchmark checks, streamed to a live AWS dashboard | Python, Lambda, DynamoDB |
| Social Media Backend API | Posts, hashtags, follows and personalized feeds on a migration-driven schema | Express, TypeORM, TypeScript |
| Task Automation API | Chained and batched task execution with retries and webhook triggers | FastAPI, asyncio |
| Trade Opportunities API | Sector-wise market reports from live search, with rate limiting and API keys | FastAPI, Gemini, SlowAPI |
| Todo App | Cross-platform task manager with filtering, sorting and offline persistence | React Native, Expo |
| MAHI AI Agent | Extracts structured fields from PDFs and answers questions over them | Streamlit, FAISS |
| AI Todo PWA | Installable offline-first todo app with service-worker caching | React, Vite, Tailwind |
| Sorting Visualizer | Five sorting algorithms visualized step by step — live | JavaScript |
| Superstore Analytics | Retail dashboard surfacing segment performance and loss drivers | Power BI, SQL |
Languages
Frontend
Backend
AI & Data
Infrastructure
Open to full-stack, backend and AI engineering roles.